Part 3/11:
A particularly controversial aspect of 01 is OpenAI's decision to hide the full reasoning chain behind the responses. Prior models like GPT-4 displayed some of their thought processes, offering transparency into how conclusions were reached. In contrast, 01 presents a filtered or concealed reasoning trail, ostensibly to maintain tighter control over the model's behavior and prevent manipulation or harm.
OpenAI explains that this approach aims to monitor and restrict risky or manipulative outputs, especially as the model's reasoning becomes more complex. While this enhances safety and alignment, it also raises concerns about transparency and trust, with curious users sometimes receiving warnings or restrictions when probing into the AI's internal logic.